Design of Gaussian process based model predictive control for seam tracking in a laser welding digital twin environment

Vision-based control systems are widely applied in laser welding, wherein the seam and weld pool can be directly monitored so that the welding speed and laser power can be adjusted. Nevertheless, seam tracking based on the coaxial vision sensor is unstable, caused by the varied illumination conditions and material surface defects. The accuracy and robustness of the controller remain a major challenge. In this study, a new seam prediction method based on Gaussian process surrogate modeling is proposed, where the sensing errors are quantified. The surrogate model predicts the future positions of the seam based on the previous observations as well as error bounds, which are utilized in the control for seam tracking. The proposed Gaussian process based model predictive control approach is evaluated in a new virtual digital environment model constructed with the globally assembled images that are captured in the real-world welding environment. The efficiency, accuracy, and robustness of the proposed method are evaluated with welding experiments. The experimental results show that the maximum tracking error in the proposed method is 42 % less than that in the standard proportional-integral-derivative controller.